Data-Driven Railway Vehicle Parameter Tuning using Markov-Chain Monte Carlo Bayesian updating

Abstract

Understanding the dynamics of the interaction between railway vehicles and tracks is essential for forecasting vehicle and track conditions and performing maintenance actions to preserve the safety of railway infrastructure. In this work, physics-based models are deployed to predict the dynamic response of railway vehicles to track alignment and irregularities. Such models comprise a large number of parameters that need to be validated and possibly tuned, a task often accomplished on the basis of expert knowledge or trial-and-error without necessarily reflecting the condition of the operating system as-is. Therefore, model uncertainties persist due to assumptions on the actual system behavior, as well as due to varying operational parameters (e.g., wheel profile, rail profile, rail moisture). To improve modelling accuracy, this study adopts a multi-body vehicle model, realized in SIMPACK software, and performs parameter tuning based on on-board measurement data (accelerations and forces) from an instrumented tilting train, which regularly traverses the Swiss Federal Railways (SBB) network. The pertinent vehicle model is optimized in terms of the interaction forces at the wheel-rail contact using a Bayesian model updating approach, relying on Markov-Chain Monte Carlo (MCMC). Specifically, the MCMC method is applied to estimate both optimal vehicle parameters and their uncertainty by minimizing the mean squared error between simulated and measured contact forces, leading to an improved vehicle twin.